Enhancing Thrust Prediction for AF-MPDTs using XGBoost Machine Learning Technique

Applied-Field Magnetoplasmadynamic Thrusters (AF-MPDTs) represent an advanced electrified propulsion system that supports the development of sustainable space propulsion. By harnessing the principles of the Lorentz force, AF-MPDTs are able to propel ionized gases to produce thrust. However, the thrust-producing mechanism is very complex and is characteristic of many coupled interactions between controllable and structural parameters. As a result, traditional analytical and empirical models for thrust prediction have faced numerous challenges over the past decades. As an alternative, this study proposes the use of machine learning (ML) technique as a substitute for traditional empirical/analytical thrust models. This paper compares the results obtained from thrust prediction by eXtreme Gradient Boosting (XGBoost) technique with results of traditional empirical models. It was observed that by carefully training XGBoost with experimental data spanning over 6 decades of work and over 58 thruster designs, XGBoost was able to predict AF-MPDTs’ thrust with 98.6% accuracy. As a result, XGBoost sets a new standard for AF-MPDT thrust prediction.